Correct some event and UMASK values according to Intel SDM,
in the Nehalem and Atom tables.
Signed-off-by: Yong Wang <yong.y.wang@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <20090609131553.GA12489@ywang-moblin2.bj.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Fill in amd_hw_cache_event_id[] with the AMD CPU specific events,
for family 0x0f, 0x10 and 0x11.
There's apparently no distinction between load and store events, so
we only fill in the load events.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Standardize and tidy up all the messages we print during
perfcounter initialization.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Fill in core2_hw_cache_event_id[] with the Atom model specific events.
The events can be used in all the tools via the -e (--event) parameter,
for example "-e l1-misses" or -"-e l2-accesses" or "-e l2-write-misses".
( Note: these are straight from the Intel manuals - not tested yet.)
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Fill in core2_hw_cache_event_id[] with the Core2 model specific events.
The events can be used in all the tools via the -e (--event) parameter,
for example "-e l1-misses" or -"-e l2-accesses" or "-e l2-write-misses".
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Extend generic event enumeration with the PERF_TYPE_HW_CACHE
method.
This is a 3-dimensional space:
{ L1-D, L1-I, L2, ITLB, DTLB, BPU } x
{ load, store, prefetch } x
{ accesses, misses }
User-space passes in the 3 coordinates and the kernel provides
a counter. (if the hardware supports that type and if the
combination makes sense.)
Combinations that make no sense produce a -EINVAL.
Combinations that are not supported by the hardware produce -ENOTSUP.
Extend the tools to deal with this, and rewrite the event symbol
parsing code with various popular aliases for the units and
access methods above. So 'l1-cache-miss' and 'l1d-read-ops' are
both valid aliases.
( x86 is supported for now, with the Nehalem event table filled in,
and with Core2 and Atom having placeholder tables. )
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Counter type is a frequently used value and we do a lot of
bit juggling by encoding and decoding it from attr->config.
Clean this up by creating a separate attr->type field.
Also clean up the various similarly complex user-space bits
all around counter attribute management.
The net improvement is significant, and it will be easier
to add a new major type (which is what triggered this cleanup).
(This changes the ABI, all tools are adapted.)
(PowerPC build-tested.)
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Throttling logic is broken and we can lock up with too small
hw sampling intervals.
Make the throttling code more robust: disable counters even
if we already disabled them.
( Also clean up whitespace damage i noticed while reading
various pieces of code related to throttling. )
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Remove the IRQ (non-NMI) handling bits as NMI will be used always.
Signed-off-by: Yong Wang <yong.y.wang@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090603051255.GA2791@ywang-moblin2.bj.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The structure isn't hw only and when I read event, I think about those
things that fall out the other end. Rename the thing.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
Cc: Stephane Eranian <eranian@googlemail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Do as Power already does, emulate sample periods up to 2^63-1 by
composing them of smaller values limited by hardware capabilities.
Only once we wrap the software period do we generate an overflow
event.
Just 10 lines of new code.
Reported-by: Stephane Eranian <eranian@googlemail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
IRQ (non-NMI) sampling is not used anymore - remove the last few bits.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
A few renames:
s/irq_period/sample_period/
s/irq_freq/sample_freq/
s/PERF_RECORD_/PERF_SAMPLE_/
s/record_type/sample_type/
And change both the new sample_type and read_format to u64.
Reported-by: Stephane Eranian <eranian@googlemail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Merge reason: merge almost-rc8 into perfcounters/core, which was -rc6
based - to pick up the latest upstream fixes.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Always use NMI for performance-monitoring interrupt as there could be
racy situations if we switch between irq and nmi mode frequently.
Signed-off-by: Yong Wang <yong.y.wang@intel.com>
LKML-Reference: <20090529052835.GA13657@ywang-moblin2.bj.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Slightly modified by trenn@suse.de -> only do this on fam 10h and fam 11h.
Currently powernow-k8 determines CPU frequency from ACPI PSS objects, but
according to AMD family 11h BKDG this frequency is just a rounded value:
"CoreFreq (MHz) = The CPU COF specified by MSRC001_00[6B:64][CpuFid]
rounded to the nearest 100 Mhz."
As a consequnce powernow-k8 reports wrong CPU frequency on some systems,
e.g. on Turion X2 Ultra:
powernow-k8: Found 1 AMD Turion(tm)X2 Ultra DualCore Mobile ZM-82
processors (2 cpu cores) (version 2.20.00)
powernow-k8: 0 : pstate 0 (2200 MHz)
powernow-k8: 1 : pstate 1 (1100 MHz)
powernow-k8: 2 : pstate 2 (600 MHz)
But this is wrong as frequency for Pstate2 is 550 MHz. x86info reports it
correctly:
#x86info -a |grep Pstate
...
Pstate-0: fid=e, did=0, vid=24 (2200MHz)
Pstate-1: fid=e, did=1, vid=30 (1100MHz)
Pstate-2: fid=e, did=2, vid=3c (550MHz) (current)
Solution is to determine the frequency directly from Pstate MSRs instead
of using rounded values from ACPI table.
Signed-off-by: Andreas Herrmann <andreas.herrmann3@amd.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Dave Jones <davej@redhat.com>
- Make the message shorter and easier to grep for
- Use printk_once instead of WARN_ONCE (functionality of these was mixed)
Signed-off-by: Thomas Renninger <trenn@suse.de>
Cc: Langsdorf, Mark <mark.langsdorf@amd.com>
Signed-off-by: Dave Jones <davej@redhat.com>
arch/x86/kernel/cpu/cpufreq/powernow-k7.c:172: warning: 'invalidate_entry' defined but not used
Reported-by: Toralf Förster <toralf.foerster@gmx.de>
Signed-off-by: Dave Jones <davej@redhat.com>
Some atom procs don't do freq scaling (such as the atom 330 on my own
littlefalls2 board). By adding the atom family here, we at least get
the benefit of passive cooling in a thermal emergency. Not sure how
to see that its actually helping any, but the driver does bind and
claim its functioning on my atom 330.
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: Dave Jones <davej@redhat.com>
We have a debug check that detects stuck NMIs and returns with
the PMU disabled in the global ctrl MSR - but i managed to trigger
a situation where this was not enough to deassert the NMI.
So clear/reset the full PMU and keep the disable count balanced when
exiting from here. This way the box produces a debug warning but
stays up and is more debuggable.
[ Impact: in case of PMU related bugs, recover more gracefully ]
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
My Nehalem box locks up in certain situations (with an
always-asserted NMI causing a lockup) if the PMU LVT
entry is programmed between NMI and IRQ mode with a
high frequency.
Standardize exlusively on NMIs instead.
[ Impact: fix lockup ]
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This reverts commit b68f1d2e7a.
It is causing problems (stuck/stuttering profiling) - when mixed
NMI and non-NMI counters are used.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090525153931.703093461@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Introduce a generic per counter interrupt throttle.
This uses the perf_counter_overflow() quick disable to throttle a specific
counter when its going too fast when a pmu->unthrottle() method is provided
which can undo the quick disable.
Power needs to implement both the quick disable and the unthrottle method.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090525153931.703093461@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
remove the x86 specific interrupt throttle
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090525153931.616671838@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Expose the INV and EDGE bits of the PMU to raw configs.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <20090525153931.494709027@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Remap percpu allocator has subtle bug when combined with page
attribute changing. Remap percpu allocator aliases PMD pages for the
first chunk and as pageattr doesn't know about the alias it ends up
updating page attributes of the original mapping thus leaving the
alises in inconsistent state which might lead to subtle data
corruption. Please read the following threads for more information:
http://thread.gmane.org/gmane.linux.kernel/835783
The following is the proposed fix which teaches pageattr about percpu
aliases.
http://thread.gmane.org/gmane.linux.kernel/837157
However, the above changes are deemed too pervasive for upstream
inclusion for 2.6.30 release, so this patch essentially disables
the remap allocator for the time being.
Signed-off-by: Tejun Heo <tj@kernel.org>
LKML-Reference: <4A1A0A27.4050301@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Introduce "noxsave" boot parameter which will disable the cpu's xsave/xrstor
capabilities. Useful for debugging and working around xsave related issues.
[ Impact: make it possible to debug problems in the field ]
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
This replaces the struct perf_counter_context in the task_struct with
a pointer to a dynamically allocated perf_counter_context struct. The
main reason for doing is this is to allow us to transfer a
perf_counter_context from one task to another when we do lazy PMU
switching in a later patch.
This has a few side-benefits: the task_struct becomes a little smaller,
we save some memory because only tasks that have perf_counters attached
get a perf_counter_context allocated for them, and we can remove the
inclusion of <linux/perf_counter.h> in sched.h, meaning that we don't
end up recompiling nearly everything whenever perf_counter.h changes.
The perf_counter_context structures are reference-counted and freed
when the last reference is dropped. A context can have references
from its task and the counters on its task. Counters can outlive the
task so it is possible that a context will be freed well after its
task has exited.
Contexts are allocated on fork if the parent had a context, or
otherwise the first time that a per-task counter is created on a task.
In the latter case, we set the context pointer in the task struct
locklessly using an atomic compare-and-exchange operation in case we
raced with some other task in creating a context for the subject task.
This also removes the task pointer from the perf_counter struct. The
task pointer was not used anywhere and would make it harder to move a
context from one task to another. Anything that needed to know which
task a counter was attached to was already using counter->ctx->task.
The __perf_counter_init_context function moves up in perf_counter.c
so that it can be called from find_get_context, and now initializes
the refcount, but is otherwise unchanged.
We were potentially calling list_del_counter twice: once from
__perf_counter_exit_task when the task exits and once from
__perf_counter_remove_from_context when the counter's fd gets closed.
This adds a check in list_del_counter so it doesn't do anything if
the counter has already been removed from the lists.
Since perf_counter_task_sched_in doesn't do anything if the task doesn't
have a context, and leaves cpuctx->task_ctx = NULL, this adds code to
__perf_install_in_context to set cpuctx->task_ctx if necessary, i.e. in
the case where the current task adds the first counter to itself and
thus creates a context for itself.
This also adds similar code to __perf_counter_enable to handle a
similar situation which can arise when the counters have been disabled
using prctl; that also leaves cpuctx->task_ctx = NULL.
[ Impact: refactor counter context management to prepare for new feature ]
Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <18966.10075.781053.231153@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
x86: DMI match for the Sony VGN-Z540N as it needs BIOS reboot,
see:
http://bugzilla.kernel.org/show_bug.cgi?id=12901
[ Impact: fix hung reboot on certain systems ]
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Cc: Len Brown <lenb@kernel.org>
LKML-Reference: <1242963350.32574.53.camel@rzhang-dt>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'tracing-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
tracing: Append prompt in /debug/tracing/README file
x86/function-graph: fix constraint for recording old return value
We have to set up the LVT entry only at counter init time, not at
every switch-in time.
There's friction between NMI and non-NMI use here - we'll probably
remove the per counter configurability of it - but until then, dont
slow down things ...
[ Impact: micro-optimization ]
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The quirk to irq_period unearthed an unrobustness we had in the
hw_counter initialization sequence: we left irq_period at 0, which
was then quirked up to 2 ... which then generated a _lot_ of
interrupts during 'perf stat' runs, slowed them down and skewed
the counter results in general.
Initialize irq_period to the maximum instead.
[ Impact: fix perf stat results ]
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Xiaohui Xin and some other folks at Intel have been looking into what's
behind the performance hit of paravirt_ops when running native.
It appears that the hit is entirely due to the paravirtualized
spinlocks introduced by:
| commit 8efcbab674
| Date: Mon Jul 7 12:07:51 2008 -0700
|
| paravirt: introduce a "lock-byte" spinlock implementation
The extra call/return in the spinlock path is somehow
causing an increase in the cycles/instruction of somewhere around 2-7%
(seems to vary quite a lot from test to test). The working theory is
that the CPU's pipeline is getting upset about the
call->call->locked-op->return->return, and seems to be failing to
speculate (though I haven't seen anything definitive about the precise
reasons). This doesn't entirely make sense, because the performance
hit is also visible on unlock and other operations which don't involve
locked instructions. But spinlock operations clearly swamp all the
other pvops operations, even though I can't imagine that they're
nearly as common (there's only a .05% increase in instructions
executed).
If I disable just the pv-spinlock calls, my tests show that pvops is
identical to non-pvops performance on native (my measurements show that
it is actually about .1% faster, but Xiaohui shows a .05% slowdown).
Summary of results, averaging 10 runs of the "mmperf" test, using a
no-pvops build as baseline:
nopv Pv-nospin Pv-spin
CPU cycles 100.00% 99.89% 102.18%
instructions 100.00% 100.10% 100.15%
CPI 100.00% 99.79% 102.03%
cache ref 100.00% 100.84% 100.28%
cache miss 100.00% 90.47% 88.56%
cache miss rate 100.00% 89.72% 88.31%
branches 100.00% 99.93% 100.04%
branch miss 100.00% 103.66% 107.72%
branch miss rt 100.00% 103.73% 107.67%
wallclock 100.00% 99.90% 102.20%
The clear effect here is that the 2% increase in CPI is
directly reflected in the final wallclock time.
(The other interesting effect is that the more ops are
out of line calls via pvops, the lower the cache access
and miss rates. Not too surprising, but it suggests that
the non-pvops kernel is over-inlined. On the flipside,
the branch misses go up correspondingly...)
So, what's the fix?
Paravirt patching turns all the pvops calls into direct calls, so
_spin_lock etc do end up having direct calls. For example, the compiler
generated code for paravirtualized _spin_lock is:
<_spin_lock+0>: mov %gs:0xb4c8,%rax
<_spin_lock+9>: incl 0xffffffffffffe044(%rax)
<_spin_lock+15>: callq *0xffffffff805a5b30
<_spin_lock+22>: retq
The indirect call will get patched to:
<_spin_lock+0>: mov %gs:0xb4c8,%rax
<_spin_lock+9>: incl 0xffffffffffffe044(%rax)
<_spin_lock+15>: callq <__ticket_spin_lock>
<_spin_lock+20>: nop; nop /* or whatever 2-byte nop */
<_spin_lock+22>: retq
One possibility is to inline _spin_lock, etc, when building an
optimised kernel (ie, when there's no spinlock/preempt
instrumentation/debugging enabled). That will remove the outer
call/return pair, returning the instruction stream to a single
call/return, which will presumably execute the same as the non-pvops
case. The downsides arel 1) it will replicate the
preempt_disable/enable code at eack lock/unlock callsite; this code is
fairly small, but not nothing; and 2) the spinlock definitions are
already a very heavily tangled mass of #ifdefs and other preprocessor
magic, and making any changes will be non-trivial.
The other obvious answer is to disable pv-spinlocks. Making them a
separate config option is fairly easy, and it would be trivial to
enable them only when Xen is enabled (as the only non-default user).
But it doesn't really address the common case of a distro build which
is going to have Xen support enabled, and leaves the open question of
whether the native performance cost of pv-spinlocks is worth the
performance improvement on a loaded Xen system (10% saving of overall
system CPU when guests block rather than spin). Still it is a
reasonable short-term workaround.
[ Impact: fix pvops performance regression when running native ]
Analysed-by: "Xin Xiaohui" <xiaohui.xin@intel.com>
Analysed-by: "Li Xin" <xin.li@intel.com>
Analysed-by: "Nakajima Jun" <jun.nakajima@intel.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Acked-by: H. Peter Anvin <hpa@zytor.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Xen-devel <xen-devel@lists.xensource.com>
LKML-Reference: <4A0B62F7.5030802@goop.org>
[ fixed the help text ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Instead of specifying the irq_period for a counter, provide a target interrupt
frequency and dynamically adapt the irq_period to match this frequency.
[ Impact: new perf-counter attribute/feature ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <20090515132018.646195868@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The treatment of the SP register is different on x86_64 and i386.
This is a regression fix that lived outside the mainline kernel from
2.6.27 to now. The regression was a result of the original merge
consolidation of the i386 and x86_64 archs to x86.
The incorrectly reported SP on i386 prevented stack tracebacks from
working correctly in gdb.
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
intel_pmu_handle_irq() can lock up in an infinite loop if the hardware
does not allow the acking of irqs. Alas, this happened in testing so
make this robust and emit a warning if it happens in the future.
Also, clean up the IRQ handlers a bit.
[ Impact: improve perfcounter irq/nmi handling robustness ]
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
On certain CPUs i have observed a stuck PMU if interval was set to
1 and NMIs were used. The PMU had PMC0 set in MSR_CORE_PERF_GLOBAL_STATUS,
but it was not possible to ack it via MSR_CORE_PERF_GLOBAL_OVF_CTRL,
and the NMI loop got stuck infinitely.
[ Impact: fix rare hangs during high perfcounter load ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Two consecutive NMIs could daze and confuse the machine when the
first would handle the overflow of both counters.
[ Impact: fix false-positive syslog messages under multi-session profiling ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The current disable/enable mechanism is:
token = hw_perf_save_disable();
...
/* do bits */
...
hw_perf_restore(token);
This works well, provided that the use nests properly. Except we don't.
x86 NMI/INT throttling has non-nested use of this, breaking things. Therefore
provide a reference counter disable/enable interface, where the first disable
disables the hardware, and the last enable enables the hardware again.
[ Impact: refactor, simplify the PMU disable/enable logic ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf_counter_unthrottle() restores throttle_ctrl, buts its never set.
Also, we fail to disable all counters when throttling.
[ Impact: fix rare stuck perf-counters when they are throttled ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
If counters are disabled globally when a perfcounter IRQ/NMI hits,
and if we throttle in that case, we'll promote the '0' value to
the next lapic IRQ and disable all perfcounters at that point,
permanently ...
Fix it.
[ Impact: fix hung perfcounters under load ]
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Take the counter width into account instead of assuming 32 bits.
In particular Nehalem has 44 bit wide counters, and all
arithmetics should happen on a 44-bit signed integer basis.
[ Impact: fix rare event imprecision, warning message on Nehalem ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
After upgrading from gcc 4.2.2 to 4.4.0, the function graph tracer broke.
Investigating, I found that in the asm that replaces the return value,
gcc was using the same register for the old value as it was for the
new value.
mov (addr), old
mov new, (addr)
But if old and new are the same register, we clobber new with old!
I first thought this was a bug in gcc 4.4.0 and reported it:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40132
Andrew Pinski responded (quickly), saying that it was correct gcc behavior
and the code needed to denote old as an "early clobber".
Instead of "=r"(old), we need "=&r"(old).
[Impact: keep function graph tracer from breaking with gcc 4.4.0 ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
